One thing that seems confusing to people when they first look at StreamInsight is that there are several development models:
1. Implicit Server
2. Explicit Server
Implicit server is fairly straightforward. You define your input stream and create a query to consume your output stream. StreamInsight does all of the heavy lifting associated with creating the server, binding your adapters, etc.
Explicit server is the most flexible but the most coding work. You specifically spin up (create and instantiate) a server object, from it spin up an application, register your input/output adapters and create them via a factory and spin up a query binding object. You use the query binding object to bind together your standing query (via a query template) and your input and output adapters. You use this query binding object to then create a query object (an instance of a query) which you then start/stop as required. With the explicit server model, you can get reuse of query templates, adapters, etc. and you can reuse an existing event stream for multiple queries. This model is also how you'd work with a server that's not hosted in-process if that's required.
IObservable/IObserver makes the creation of an input stream quite easy. It is based on the RX extensions (Reactive Extensions) to the .NET Framework. You take an enumerable class (ie: one that implements IEnumerable) and you create an Observable class from it. You then hook up subscribers to the observable class.
Picking the right model is something you'll need to do fairly early on in your StreamInsight project.